Image processing generally refers to the manipulation of pictorial data using computers. Computer Aided Design (CAD) is one example of how computers are used to draft complex engineering drawings such as mechanical, architectural or electrical drawings. Other examples of image processing include the manipulation of still photographs or cinema to achieve various effects such as feature enhancement, three-dimensional rendering, or animation. However, the term "image processing" is not limited to pictures--it generally refers to the digitization and computer processing of any analog signal that represents something physical and perceptible in the world. An audio signal can be digitized and processed with computers to perform manipulations such as noise reduction or voice recognition; modulated electrical signals, such as telecommunications or cardiograms, can be digitized and processed by computers to extract pertinent information, and so on. The common factor in image processing is the need to digitize the continuous-time analog signal into a discrete-time digital signal that can be operated on by computers using boolean mathematics implemented in transistor logic circuits.
A common format for storing and manipulating digitized signals is simply to represent the digital data as a bit-map of pixels, where each pixel represents a particular characteristic (e.g., magnitude or color) of the analog signal. Typically, a picture is digitized into a bit-map of individual pixels where each pixel represents a color of the image in a very small, localized area. When the pixels are displayed as a congregation, the original picture appears without loss of perception to the human eye as long as there is sufficient resolution (number of pixels) both in spatial and color contexts. A black and white picture can be represented with one-bit pixels, where the state of the bit (0 or 1) represents the two colors, black and white. To digitize a picture comprising multiple colors, each pixel is represented using n-bits such that each pixel can take on one of 2.sup.n different colors. The process of converting an analog image into a bit-map image is referred to as "rasterizing" the image.
There are several well known problems with raster images that limit or inhibit the full potential of computer manipulation. The amount of memory necessary to store a large photograph having many colors (thousands or even millions of colors) can be immense, and the problem is exacerbated when attempting to digitize a series of photographs such as a scene in a movie. Not only do raster images require large amounts of storage memory, but processing such a large amount of data can be slow, particularly when attempting to transfer raster images over a network such as the internet.
Another problem inherent with the raster format is an inability to perform many of the desired image processing functions, such as three-dimensional rendering or animation. Even a simple operation such as magnifying the image results in a granular distortion caused by enlarging the pixels of the image as illustrated in FIG. 1. It is possible to perform digital signal processing when magnifying a raster image by effectively re-sampling the image to attenuate the granular distortion that occurs due to the fixed resolution. However, this process only reduces the granular distortion rather than eliminate it. Furthermore, resampling the image is time consuming and not practical in many applications. Other operations, such as electronic focusing and automatic imaging, are simply not possible when the image is stored in a raster format.
The alternative is to convert the raster image into a mathematical format known as vectors. A vector is a mathematical description of the image which is not constrained by a fixed resolution as with raster data. Furthermore, vectors allow for a much more diverse range of image manipulations due to the mathematical representation of the image. A simple example of a vector is a line beginning at a particular X,Y point in a Cartesian plane and ending at another X,Y point in that plane. To date, however, converting raster images into a series of linear vectors has been largely unsuccessful. This is due to the immense number of different possible patterns that raster images can form. For example, a 3 by 3 pixel matrix of a nine color picture can take on about 21,000 different possible patterns. Each possibility must be accounted for when converting the image into a set of linear vectors. If just one more pixel is added to the matrix, the complexity increases exponentially. Consequently, techniques for converting raster images into linear vectors normally operate on only a small section of the image at a time. The problem with this approach is that the resulting vectorized image appears fragmented due to discontinuities when the individual sections are combined.
There have been other attempts to convert a raster image using vectors that are more abstract mathematical representations. For example, a technique referred to as wavelets attempts to represent the raster image as a series of interconnected mathematical equations. With wavelets, a particular feature of the image might be represented using a number of polynomials which approximate the feature contours. The problem with this technique, however, is that the equations become extremely complex unless the operation is constrained to only a small group of pixels. But again, distortion due to fragmentation occurs when attempting to combine the individual equations into a complete image. A similar distortion occurs with using fractals, a method where the image is represented using a library of different shapes which approximate the various features of an image. Similar to wavelets, however, discontinuities occur with fractals when attempting to combine the individual shapes which causes distortion due to fragmentation.
Another known method for raster to vector conversion, referred to as the Automatic Bezier Curve technique, draws Bezier curves through tags that are placed somewhat arbitrarily on the image. Unlike fractals or wavelets, Bezier techniques cannot convert complicated pictures without losing information. On the other hand, the Bezier technique can sequence equations together over a large number of pixels which reduces the amount of fragmentation typically associated with other techniques.
The Bezier method is illustrated in FIG. 1B. The first step is to identify the color borders for each feature in the image, and then to place somewhat arbitrarily tags on the borders. These tags then become the new representation of the image. When the image is to be displayed, these tags are converted into a rasterized format by drawing Bezier curves through the tags as shown in FIG. lB. The problem with the Bezier conversion technique is that it has a hard time dealing with complicated images. When tags are placed some definition is lost. In a complex image, many tags are placed resulting in a great deal of loss in picture definition. Even more severe is that the Bezier technique only looks at simple borders. In complex images, there is usually an interaction of shades that catch the eye. The Bezier curve technique treats all of these different shades as either borders or points. This often results in a vector representation that is more complicated than a raster representation; in effect, the Bezier solution can actually be worse than the problem.
There have been other attempts to represent the feature contours of an image using mathematics capable of spanning more than just a small area on the image. For example, U.S. Pat. No. #4,777,651 discloses a method of pixel-to-vector conversion where the line and edge features of the image are converted into a series of continous line segments. The line segments are generated by connecting a series of slope elements which follow the contour of a line or edge in the image. The slope elements are generated by scanning a 3.times.3 array of pixels through the raster image data and generating a slope vector for each pixel. The 3.times.3 array of pixels are evaluated to determine the slope of the image contour at the current pixel. The slope elements are then connected to form continous line segments that represent the contours of the image. The line segments are further processed to generate the vectors that can be understood and manipulated using computers.
The process of sequencing a series of slope elements into line segments which represent the contours of an image has been known in the art as early as 1961. See H. Freeman, "On the Encoding of Arbitrary Geometric Configurations," IRE Transactions, EC-10(2), June 1961, 260-268. Freeman describes the technique of scanning a square matrix through raster data to generate a slope element for each pixel, and then connecting the slope elements into line segments which represent the image contours.
There are drawbacks with the aforementioned prior art methods of sequencing slope elements into line segments to represent the image contours. Namely, U.S. Pat. No. 4,777,651 is capable of sequencing line segments for contours of only black and white images. Thus, this technique cannot be applied to color images without first converting the image into black and white which means the color information is lost. Furthermore, the list of square matrices for generating the slope elements shown in FIG. 9 is not exhaustive and, consequently, fragmentation can occur when attempting to generate and connect the line segments. Still further, sequencing the line segments using only the slope values does not provide optimal information for generating the final vectors which ultimately represent the image. In other words, the vectors generated using this technique are not optimal because there is not enough information provided in line segments comprised of only slope values. Also, these techniques do not address the optical illusion which can manifest in images where the contrast in certain features determines or enables perception.
There is, therefore, a need for an improved method and apparatus for converting the contours or borders of a raster image into a more abstract mathematical representation which can be converted into a vector format without fragmenting the image. More specifically, it is an object of the present invention to convert the color borders of a raster image into a mathematical representation, wherein the raster image is comprised of pixels representing more colors than just black and white. Another object of the present invention is to provide enough information in the border representation of the image so as to optimize the vectors ultimately generated from this information. Still another object of the present invention is to identify and resolve contrast conflicts in the various features of an image in order to avoid misperceiving the vectorized image when converted back into a raster format.